I asked chatgpt and it told me how to split my typescript project into multiple smaller ones. I use Zod quite a bit and it made 250k "instantiations" ... Now it is only 80k max per project.. much better ππ
@jerome-etienne
I am a #SaaS solo founder and currently building a great startup https://ContentMagick.ai . It produces your social media 10x faster thanks to #AI π§ β¨ Free tier available! No credit card required! π Bsky: @contentMagick.ai #buildInPublic #soloPreneur
I asked chatgpt and it told me how to split my typescript project into multiple smaller ones. I use Zod quite a bit and it made 250k "instantiations" ... Now it is only 80k max per project.. much better ππ
My typescript server was slow in vsCode... This makes everything so slow, seconds to paste 10 characters, seconds to save the file... Horrible. So I had to stop for 30 min the time to figure out how to make it usable for again. #sluggishCoding #buildInPublic
Strange token usage on #OpenAi . Here is the response usage for the same request on gpt-5.2, gpt-5.mini and gpt-5-nano ... The number of tokens varie widely (up to x20)... which make gpt-5-mini 2.65 time more expansive than 5.2 ... This can not be right ...
ai video project im working on #buildInPublic
@monde-des-licornes.bsky.social hey love what you do? Could we have a chat? J'ai vu que tu as pris une pause de 8 ans sur youtube et que tu as recommencΓ© cette annΓ©e.. je pense a contribuΓ© dans le mΓͺme vibre que les monde des licornes, j'aimerais Γ©changer. Je suis sΓ©rieux, rΓ©ponds stp ππ
Just added per-word transcription to my videos in minutes with Remotion + whisper.cpp π₯
www.remotion.dev/docs/install...
Powered by the open-source Whisper model from OpenAI.
Local, fast, word-level timing. #Remotion #AIvideo
character sheet in the wild... or how to include your characters on a given set. Here are 2 different sets, notice how the characters look the same, this means character sheets are working π #AI #buildInPublic
How to keep character consistency and personality stable over time ? TIPS: use character sheets π #AI #buildInPublic
OpenAI voices lack robust user controls. While there is an 'instructions' field meant to guide the emotional tone, the model mostly ignores it. They are great for reading text, but fall short when it comes to actual voice acting. #AI #buildInPublic
More #AI generated content. Using character sheet. I start to get a better controls over it π #builtInPublic
Having fun with AI generated video and remotion. #early #builtInPublic
huber loss is so much better than mae!!! #chess #BuildInPublic #ai #DataScience#DeepLearning
vscode got a surprisingly good pytorch support !!
#chess #ML #buildInPublic #AI #MachineLearning #DeepLearning #DataScience #AIDev #Tech #Python #OpenSource
code.visualstudio.com/docs/datasci...
I'm diving into some serious hyperparameter tuning for my #chess program! βοΈ
I'm using Optuna, and seriously, what a great APIβit's super easy to use and makes the process so much cleaner. Getting closer to finding that optimal configuration!
#buildInPublic #chess
Fun one for python user: "Where is the bug ?"
The following code crashes
Little riddle: "The following code crashes, where is the bug ?"
Definitely! it makes refactoring so much easier too!!
Little bug in vscode python. For those unfamiliar to python `1` is completly different from `1.0`. The former is a int, and the later is float. No big deal, but confusing π΅βπ«
Progress in my #chess engine based on machine learning. Look at those curves! The way the validation follows the training, they feel like that made to go together π #buildInPublic
Level Unlocked! π Spent the night diving deep into the UCI Protocol (Universal Chess Interface). This is the secret handshakeβthe standard that lets you chat with any #chess engine via stdin/stdout. Suddenly, building my own engine feels a whole lot more real! Standard achieved. π #buildInPublic
im vibe coding alpha zero!!! #chess #buildInPublic
A neural network model on its own cannot produce a reasonable move, as it lacks the depth of calculation required. #chess #buildInPublic
#chess fun, using fishtest pgn files (aka the chess game done by stockfish during testing), i found a "+M77" which means stockfish is evaluating that it will do Mate in 77 moves.... This is depth in the sea of chess π€ͺπ³β #buildInPublic
OK tonight im learning the FEN specs #chess #builtInPublic en.wikipedia.org/wiki/Forsyth...
Coding a #chess in pytorch... how come there is no early stopper in pytorch... everybody has to recode its own... #buildInPublic
Raise your hand if you code enough #chess to be able to be able to encode a board in FEN by hand πββοΈπ
Fun in #Chess π #Stockfish is crashing when you send it a position which is illegal (e.g. invalid castling rights)... and i means crashing, not returning an error... Gasp π
For those interested, it is all happening on this repo!
github.com/jeromeetienn...
Progress in my #chess model π, real happy about the result π
Note how the training loss is above the validation loss, this is due to a nice balance between the learning rate scheduler and the dropout probability. This means it fight against overfitting efficiently β€οΈ
More on chess via ML tomorrow!
Tonight i learned Monte Carlo Tree Search (MCTS), a nice way to handle AI, where AI doesn't understand the game it is playing, it just play it randomly... More powerfull that it seems at first π
I coded a tictactoe, connect4 and othello in the everning. It is that fast.
github.com/jeromeetienn...